C
DEMAND FORECASTING
IN MARKETING *
An important part of the marketing planning process is the setting of goals that are
realistic and achievable, given a particular marketing environment and level of marketing
commitment. In marketing, such goals are usually based on market share objectives and
sales targets, both of which require accurate forecasts of total market size, market size of
target segments and likely market share within a targeted segment.
W H AT A R E W E F O R E C A S T I N G ?
Accurate forecasting requires a clear definition of the market in question. Markets may be
differentiated on the basis of the following variables.
GEOGRAPHY
A market may be defined at world, country, state, region, sales territory, town, store or
customer level. When formulating a forecast or other marketing plans, the geographical
dimension must be clearly indicated. Planning Coca Cola consumption for the year 2000
Sydney Olympics for example, will necessitate the forecasting of increased consumption
for the Sydney sales region, but not necessarily for Brisbane.
TIME
A forecast must be defined for a specific time period. Initially several levels of forecast
may be set at differing levels of specificity for the short term, the medium term and the
long term, with increasingly larger confidence intervals around the forecasts, the further
into the future the projection. So for the next year company X may forecast its sales of
product B as being $2.1 million, with a 95 per cent confidence limit of + or −$0.3 million.
However its long range forecast, say for the year 2010 might be $13.5 million, with a
95 per cent confidence interval of + or −$1.8 million. Generally companies will set specific
forecasts on a monthly, quarterly or annual basis.
PRODUCT LEVEL
A forecast may be set for the industry, the company and the product. Even at product
level, separate forecasts might be made for product assortment, product line and product
item. Johnson and Johnson might use survey analysis and demographic data to forecast
the industry sales of the product assortment, shampoo, for the forthcoming year; from this
total industry figure the company might further estimate its market share to create a
company forecast. From this forecast, using past sales projections and company sales
strategy information, Johnson and Johnson might further dissect a forecast for the
product line of children’s shampoo. A further dissection might produce a forecast for
the product item No More Tears in a 500 ml bottle.
1. wants to buy your assortment, line and item of product and has not
already satisfied their need for this demand in the current time period
2. has the income to pay for your product
3. is interested in buying the brand which your company has for sale.
INDUSTRY FORECAST
An industry forecast considers not only the overall state of the economy but also factors
specific to the industry in which a company is operating. In the case of CSR Building
Materials, management will not only be interested in the overall state of the economy, but
also in the state of the building sector specifically. Some economic effects might be lagged,
for example, so that their positive or negative effects on a particular segment might be
considerably delayed. This is often the case with changes in interest rates, which might
affect the number of new housing starts but not building in progress. In developing an
industry forecast, consideration must also be given to the level of competition, both
within the particular industry and confronting the industry from the outside in the form
of substitute products or foreign imports. Whilst a marketer generally does not have to
develop forecasts for an entire industry, the onus is on he or she to have a clear under-
standing of the parameters of the industry in which he or she is operating.
* * *
Stem-and-Leaf Plot
Forecasting based on time-series uses the company’s historical sales data to search for
certain patterns which are projected forward in time to produce a future forecast. There
are four components in this type of analysis:
1. Trend describes the underlying growth or decline pattern in the sales. This is
the major thrust of sales progress. It may also reflect strong underlying factors,
such as economic, demographic or technological trends. A linear trend is the
simplest trend to understand mathematically. However, non-linear forms such
as quadratic or exponential trends may also be estimated using mathematical
models and time-series data.
▼
692 Appendix C
Examples of trends
(a) Linear—for example, a product experiencing a consistent rate of growth, or decline.
2. Season refers to movements in a series which occur at regular times over the
years. This pattern of movement in the data may be detected in quarterly,
monthly, daily or even hourly data and may be related to the weather or
climatic seasons, to patterns of business activity (such as a flurry of activity at
the end of June because of the end of the financial year) or to patterns of
consumer behaviour (such as increased consumption of toys at Christmas).
3. Cycle captures the longer swings up and down which are often caused by
economic upswings and downturns, and similar cyclical swings in consumer
confidence, consumer tastes, fads and competitive effects. Cyclical effects are
difficult to predict because they do not recur on a regular basis. However,
approximate cyclical effects can be built into medium- and long-range forecasts.
Time-series analysis involves defining the data series components of trend, cycle, TIME-SERIES ANALYSIS
season and irregularity and combining them to form a sales forecast. This technique is Defining the data series
particularly effective for forecasting sales of products with relatively stable demand. Time components of trend, cycle,
series analysis treats sales as being a function of time which in fact acts as a proxy for a season and irregularity and
multitude of other variables, such as competition, consumer preferences or economic combining them to form
activity. a sales forecast
Sometimes, however, we may prefer to deal with these variables directly in making
our forecast. Relationship methods enable us to do just this.
RELATIONSHIP METHODS
Leading indicators The demand for many products is derived from some other activity, LEADING INDICATORS
the statistics for which can provide the basis of a forecast for the product in question. For Statistics which can provide
example, CSR Building Materials has much of its demand derived from housing starts and the basis of a forecast for the
building approvals, which in turn are dependent to a large extent on the health of the product in question
economy. Many leading indicators are supplied to the building industry through the
Housing Industry Association, the Indicative Planning Association and BIS Shrapnel.
Economic indicators such as the Consumer Price Index (CPI) and the All-ordinaries Share
Index are also available for use in forecasting.
Regression analysis This enables us to express sales as a variable which is affected REGRESSION ANALYSIS
by (or dependent on) one or more other (independent) variables. We might conclude, for Analysis which allows for the
example, that demand for CSR bricks will be at least partly dependent on the number of expression of sales as a variable
housing starts in some preceding periods. which is affected by one or
Although we mentioned numerous types of trends in the previous section, we will more variables
concentrate here only on the linear trend.
▼
694 Appendix C
Notation:
The dependent variable, e.g. sales, is usually denoted by Y
The independent variables are denoted by X1, X2...Xn
Y * denotes the ‘fitted’ Y values, when the original X values are substituted into the
regression equation
In statistical terminology we say that sales is a function of one or more independent
variables:
Y = f( X1, X2...Xn )
Simple linear regression refers to a regression model with only one independent
variable. Multiple regression models have more than one independent variable.
SIMPLE REGRESSION MODEL We will discuss here only the simple regression model.
A regression model with only
one independent variable The form of the simple linear equation is:
Y * = a + b 1X 1
where:a = the intercept term (the point at which the line crosses the Y axis or
where X = 0)
b = the slope coefficient. As X increases 1 unit, Y * increases by b units.
If b is negative, the equation will be that of a negatively-sloped line—that is,
as X increases, Y * will decrease.
Figure 1 Y
A theoretical regression
line showing the role of Y * = a + bX
the slope and intercept
b
a
X=0 1 unit X
Formulae for estimating the simple linear regression equation are as follows:
Example: Assume that the sales of a particular type of CSR roof tile is dependent on
the number of housing starts in the preceding half year. Other factors of potential
importance will be ignored in this simple equation. The following (fictitious) data shown
in Figure 2 might be considered, and are represented graphically in Figure 3.
Figure 2
Hypothetical data for PERIOD X = HOUSING STARTS IN Y = SALES OF CSR
CSR roof tile sales PRECEDING HALF YEAR ROOF TILES
(THOUSANDS) (MILLIONS)
1 100 43
2 112 54
3 165 91
4 150 80
5 121 60
▼
Demand Forecasting in Mar keting 695
Y Figure 3
60 x
x
40 x
X
100 120 140 150 160
Figure 4
PERIOD X X – X (X – X)2 Y Y – Y (X–X) (Y–Y) Intermediate regression
calculation
1 100 –29.6 876.16 43 –22.6 668.96
2 112 –17.6 309.76 54 –11.6 204.16
3 165 35.4 253.16 91 25.4 899.16
4 150 20.4 416.16 80 14.4 293.76
5 121 –8.6 73.96 60 –5.6 48.16
a = Y – bX = 65.6 – (0.7218)129.6
= 65.6 – 93.55
= –27.95
This means that for every 1 unit of X (every 1000 housing starts), sales of roof tiles will
= 0.7218 units of Y (0.7218 x 1 million roof tiles = 721 800 roof tiles).
a = –27.95 has a less significant meaning, being negative. This sometimes happens in
a mathematical regression equation, with a positively-sloped line.
This is an example based on fictitious data, but it serves to show the workings of
a regression equation. In the real world of business, the tedium of such calculation
is removed by the use of statistical and forecasting computer software, such as SPSS,
Microtab, TSP and Forecast-Pro.
▼
696 Appendix C
Many more statistics can be calculated in relation to regression analysis, which enables
hypotheses to be tested relating to the importance of variables used in the equation and
to the overall value of the equation as a whole. However, as with any complex statistical
methodology, caution must be used in the design, conduct and analysis, as incorrect initial
assumptions can invalidate the entire process. Nonetheless, in the hands of experienced
and knowledgeable users, and with the use of increasingly available and sophisticated
computer technology, regression techniques provide a useful quantitative basis for many
business forecasts. Further details of regression analysis can be obtained from statistics or
econometric textbooks.
3. The final column of Figure 5 shows the index (sales/CMA). Values below 100
show that sales are less than the centred moving average, and vice versa. Figure 6
summarises the index column from Figure 5, and calculates the average index for
each of the respective quarters. A final adjustment ensures that all four indices
add up to 100. A quick check of the dimensions of these seasonal indices con-
firms the expected seasonal pattern of champagne sales, with the highest index
value occurring in quarter IV—the festive season.
Figure 5
KEY MT = moving total
Fictional data of sales of a
MA = moving average champagne product in A$
CMA = centred moving average million.
SI = seasonal index
T = trend
Figure 6
Quarter I II III IV Calculation of seasonal
index (using index column
91.80 141.54 from Figure 5)
81.16 78.95 50.00 151.96
66.35 81.44 97.87 139.00
68.03 96.57
Figure 7
Calculation of regression trend
Calculation of regression
trend for champagne
Let Y = Sales data as above sales forecast
Let X = Time from 1 to 16
The interpretation of the b value of 0.1738 is that for every increase in time period, sales
will increase by 0.1738 million dollars or $173,800.
▼
698 Appendix C
Let us now assume that we wish to forecast two years into the future beyond the
given data. Given that our data stops at period 16, we will wish to be forecasting sales for
periods 17 to 24. Thus in the equation Y* = 0.7787 + 0.1738t, we will be substituting pro-
gressively the values t = 17, t = 18...t = 24 to get the trend base of the forecast. We will
then multiply this by the seasonal index to obtain the final forecast.
Figure 8
Combining regression PERIOD QUARTER TREND SEASONAL FINAL FORECAST
trend and seasonal indices NUMBER VALUE INDEX/100 TXSI
to create final forecast
17 I 3.73 0.7533 2.81
18 II 3.91 0.8979 3.51
19 III 4.08 0.8375 3.42
20 IV 4.25 1.5113 6.42
21 I 4.43 0.7533 3.34
22 II 4.60 0.8979 4.13
23 III 4.78 0.8375 4.00
24 IV 4.95 1.5113 6.80
It is important to note that this method presupposes that a sufficiently long sales his-
tory exists to be able to develop a seasonal index, and that pre-existing conditions will
hold for the next eight periods. In reality this may not be so, but this statistical forecast
base can then be adjusted for the findings of qualitative research. The overlay of a cycli-
cal effect may also be necessary. However, despite the fact that such techniques might
have some shortcomings, they are usually much more satisfactory than pure ‘guessti-
mates’. In practice, adjustments to forecasts obtained through quantitative methods may
be made on the basis of more qualitative methods. Some of these will now be discussed.
EXPERT OPINION
At times, forecasts may be made with little formal statistical analysis, relying instead on
the use of experienced judgment by executives, marketing consultants, trade associations,
buyers/sellers or academics.
Forecasts made by executives may be based on intuition rather than scientific
methodology. Their success probably depends on the degree of insight into the overall
market behaviour of the particular executive. In a large organisation, reliance on such
judgment without the backup of more rigorous methods may prove to be risky in the
long term, especially in a fluid competitive environment.
Figure 9
Sales in $million
Graphic representation of
original sales, and forecast 6
Actual
➤
➤
4
➤
➤
2 Forecast